Research map / 3D vision / 3D generation
Order-Aware Generative Modeling Using the 3D-Craft Dataset
Zhuoyuan Chen, Demi Guo, Tong Xiao, Saining Xie, Xinlei Chen, Haonan Yu, Jonathan Gray, Kavya Srinet, Haoqi Fan, Jerry Ma, Charles R. Qi, Shubham Tulsiani and 2 more
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| learn from natural human building order instead of raster-scan order (VoxelCNN vs Learned Raster-Scan) | VoxelCNN · 3D-Craft test | ACC@1 (%): 31.5 -> 62.7 | Improved |
| model spatial structure with 3D convolutions instead of an LSTM over action embeddings | VoxelCNN · 3D-Craft test | ACC@1 (%): 32.1 -> 62.7 | Improved |
| use VoxelCNN rather than Nearest Neighbor look-up of occupancy patches | next-block prediction · 3D-Craft test | ACC@10 (%): 61.9 -> 78.9 | Improved |
| use VoxelCNN rather than naive inertia heuristic | next-block prediction · 3D-Craft test | MTC: 287.7 -> 122.8 | Improved |
| LSTM instead of simple heuristic (Naive Inertia) | sequential baseline · 3D-Craft test | ACC@1 (%): 38.7 -> 32.1 | Hurt |
| use human-order VoxelCNN rather than 3D PixelCNN raster-scan | order-aware generation · 3D-Craft test | Perplexity: 3.79 -> 3.24 | Improved |
| give the model the final object as extra input for order recovery | VoxelCNN · 3D-Craft test | ACC@1 (%): 62.7 -> 69.3 | Improved |
| give the final object to Nearest Neighbor in order recovery | Nearest Neighbor · 3D-Craft test | ACC@1 (%): 42.9 -> 43.3 | No clear effect |
Closest papers by meaning
- Rigel3D: Rig-aware Latents for Animation-Ready 3D Asset Generation (NeurIPS 2026 · similarity 0.86)
- 3DCodeBench: Benchmarking Agentic Procedural 3D Modeling Via Code (NeurIPS 2026 · similarity 0.85)
- 3D-VLA: A 3D Vision-Language-Action Generative World Model (ICML 2024 · similarity 0.85)
- Puppeteer: Rig and Animate Your 3D Models (NeurIPS 2025 · similarity 0.85)
- CrafterDojo: A Suite of Foundation Models for Building Open-Ended Embodied Agents in Crafter (NeurIPS 2026 · similarity 0.84)
- DreamCraft3D: Hierarchical 3D Generation with Bootstrapped Diffusion Prior (ICLR 2024 · similarity 0.84)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
